Architectures for Adaptive Low-Power Embedded Multimedia System
Muhammad Shafique (Universität Karlsruhe, KIT)
The extreme complexity/energy requirements and context-aware processing nature of next generation multimedia applications stimulate the need for adaptive low-power embedded multimedia systems with high-performance. Run-time adaptivity is required to react to the run-time varying scenarios (e.g., quality and performance constraints, available energy, input data). This talk accentuates the basic shortcomings of state-of-the-art and discusses the emerging trend of dynamically reconfigurable processors along with their energy-related issues, i.e., reconfiguration power, leakage power, and lack of efficient energy management features. Afterwards, novel techniques for adaptive energy management at both processor architecture and application architecture levels are introduced that exploit the available potential of energy reduction in adaptive multimedia systems (based on dynamically reconfigurable processors) while meeting the performance/area constraints and keeping the video quality degradation unnoticeable, under run-time varying scenarios. An emphasis is placed on the fact that power-energy reduction needs to be combated at various abstraction levels. At the processor architecture level the novel concept of Selective Instruction Set Muting is introduced that raises the abstraction level of power-shutdown to the custom instruction set level, i.e., an instruction-set oriented shutdown. It thereby enables a far higher potential for leakage energy savings. A run-time adaptive energy management scheme determines an energy-minimizing instruction set and performs the Selective Instruction Set Muting on the temporarily unused instruction set in order to minimize the overall energy. This talk additionally discusses the application-level adaptivity and energy reduction using an advanced video encoder (like H.264/AVC) that incorporates the novel concept of Energy-Quality Classes, video properties dependent adaptive complexity reduction, and an adaptive energy budgeting scheme to realize the adaptive low-power video encoding. Compared to state-of-the-art, the proposed contribution provides significant energy savings. Altogether, the proposed processor and application architectures enable adaptive embedded multimedia systems with low power/energy consumption to provide means for next-generation mobile multimedia applications and emerging multimedia standards.
Muhammad Shafique has a 7+ years' research and development experience in adaptive, low-power, and high-performance embedded multimedia systems in leading industrial and research organizations. He received his B.Sc. Engineering degree with 4 Gold Medals and M.Sc. Information Technology degree with 2 Gold Medals from Pakistan. Since 2006, Mr. Shafique is pursuing his Ph.D. with Prof. Dr. Jörg Henkel, Karlsruhe Institute of Technology (KIT), Germany. His main research interests are extensible processors and reconfigurable computing systems for adaptive low-power embedded multimedia with a focus on concepts that allow systems to dynamically adapt to the run-time changing requirements/situations. Mr. Shafique holds one US patent, DATE'08 Best Paper Award, ICCAD'10 Best Paper Nomination, HiPEAC Paper Award, and Best Master's Thesis Award. He has 20+ research publications in the premier conferences and journals. Mr. Shafique has also served as an external reviewer for various premier conferences and journals (including DAC, ICCAD, DATE, TVLSI, TCSVT, etc.). He is a member of IEEE and ITG/VDE.